ROAD2H: Development and evaluation of an open-source explainable artificial intelligence approach for managing...

Area of Science:

• Health Informatics
• Artificial Intelligence in Medicine
• Clinical Decision Support Systems

Background:

• Clinical decision support systems (CDSSs) require integration of clinical guidelines to address patient comorbidities and potential conflicts.
• Existing systems often lack transparency and explainability, hindering adoption in complex patient cases.
• Middle-income countries face unique challenges in deploying advanced healthcare technologies like electronic health records (EHRs).

Purpose of the Study:

• To develop and evaluate a non-proprietary, standards-based approach for deploying computable guidelines with explainable argumentation.
• To integrate this system with a commercial EHR in Serbia, focusing on chronic obstructive pulmonary disease (COPD) with comorbidities.
• To assess the system's accuracy and usability through simulated cases and expert review.

Main Methods:

• Utilized an ontological framework (Transition-based Medical Recommendation - TMR) and Assumptions-Based Argumentation with preferences and Goals (ABA+G) for guideline representation and conflict mitigation.
• Implemented remote EHR integration using a microservice architecture based on HL7 FHIR and CDS Hooks.
• Developed a prototype for managing COPD with comorbid cardiovascular or chronic kidney diseases, evaluated with 20 simulated cases and five pulmonologists.

Main Results:

• Pulmonologists achieved 97% agreement with the CDSS's COPD symptom severity assessment and 98% agreement with proposed care plans.
• Experts favored the explainable argumentation principles of the system.
• Suggestions included incorporating additional comorbidities and customizing explanation levels based on expertise.

Conclusions:

• An ontological model offers a flexible approach to providing explainable AI and argumentation for managing long-term conditions.
• The developed system demonstrates high accuracy and expert acceptance in a real-world clinical context.
• Further validation is needed by extending the approach to other guidelines and multiple comorbidities.